Evaluation of Propensity Score-Calibration and Multiple Imputation for Missing Confounder Data in EHR-Based Comparative Effectiveness Research

• Activity Number: 312 - What We Know About What We Don’t Know: Overcoming Incomplete Data in Practice
• Type: Invited
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Rebecca Hubbard* and Daniel Vader and Ronac Mamtani
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
• Keywords: missing data; measurement error; propensity score; real-world data; EHR; comparative effectiveness

Abstract

Electronic health records (EHR)-derived data represent an enormous research resource with exposures and outcomes for a large and diverse population. However, EHR data have many limitations including measurement error and missing data. Comparative effectiveness research (CER) conducted using EHR often seeks to estimate target parameters using inverse probability of treatment weighting (IPTW). In this context, missingness in confounders can be handled in multiple ways. Multiple imputation (MI) can be used to impute values for confounders with missingness prior to estimation of the propensity score. Alternatively, propensity score calibration (PSC) transforms this missing data problem into a measurement error problem. The PSC approach has potential to alleviate the computational burden of MI in large EHR databases. Motivated by a real-world study of treatments for bladder cancer using oncology EHR data, we used plasmode simulation to characterize performance of IPTW hazard ratio estimates accounting for missingness using MI or PSC. We provide guidance on when these approaches can be used to obtain valid inference in CER using EHR data when information on confounders is incomplete.